A Multidimensional Examination of Jury Composition,
Trial Outcomes, and Attorney Preferences∗
Jee-Yeon K. Lehmann†
University of Houston
Jeremy Blair Smith‡
Boston University
June 27, 2013
Abstract
We assess the degree to which seated juries in U.S. criminal trials might fall short
of the constitutional ideal of impartiality. We first ask if certain demographic and socioeconomic characteristics are related to pre-deliberation biases that individual jurors
hold or to the verdicts at which juries arrive collectively. We do not focus solely on
race, but also jointly consider other characteristics – sex, age, religiousness, education,
and income – that existing literature has largely neglected. A uniquely rich dataset on
non-capital felony jury trials held in four major state trial courts allows us to identify
within-case effects and to control for typically unobservable aspects of the trial and
its participants. We find that jurors with higher income and religiousness hold more
favorable sentiments for the prosecution, while blacks hold more favorable sentiments
for the defense. These pre-deliberation biases are reflected in trial outcomes, with juries
with a higher average income and a greater proportion of religious jurors acquitting on
fewer counts, and juries with a greater proportion of blacks convicting on fewer counts.
However, these jury composition effects are smaller and account for less of the explained
variation in verdicts than the effect of evidentiary strength. Moreover, while we find
that prosecuting/defense attorneys prefer juries with higher/lower average income, indicating that attorneys are aware of the effect of income on predispositions and verdicts
and may therefore attempt to leverage this knowledge to manipulate trial outcomes in
their favor, we also find evidence that they are mistaken about the effects of other
characteristics. Our results thus raise some concerns regarding the trustworthiness of
U.S. criminal trials, but also provide important context for such concerns, especially by
illustrating that the sources of jury bias may be more nuanced and multidimensional
than an analysis based on race alone would imply.
JEL Codes: K4, J7
∗
We are especially grateful to Kevin Lang, Claudia Olivetti, and Michael Manove for their guidance and
encouragement on this project. Special thanks to Johannes Schmieder, Marc Rysman, Daniele Paserman,
Aimee Chin, Steven Craig, Scott Imberman, Chinhui Juhn, Jungmo Yoon, Andy Zuppann, Michael Luca,
David Rossman, and Hsueh-Ling Huynh for helpful comments. Thanks as well to the many other colleagues
and seminar participants who have offered feedback along the way. Any errors are our own.
†
E-mail: jlehmann@uh.edu. This material is partly based upon graduate work supported under the
Department of Education Jacob K. Javits Fellowship, the Mustard Seed Foundation Harvey Fellowship, and
the Boston University Tuition Fellowship.
‡
E-mail: jersmith@bu.edu. I would like to acknowledge financial support from the Social Sciences and
Humanities Research Council of Canada and the Boston University Department of Economics.
1
Jury Composition, Trial Outcomes, & Attorney Preferences
1
Lehmann & Smith
Introduction
Legal scholars and the public have frequently voiced concerns that the American legal system
fails to provide defendants with their constitutional right to an impartial jury in criminal
prosecutions.1 The focal point of such concerns is typically the jury selection process, which
is regarded as giving attorneys a means to stack juries with biased jurors.2 But underlying
this view is a supposition that attorneys can identify such biased jurors in the first place.
This identification is straightforward if juror biases are correlated with observable juror
characteristics such as race, which they are commonly assumed to be. Thus, the trial of the
police officers charged with the beating of Rodney King – in which an entirely non-black
jury failed to convict four white men of excessive force against a black victim despite visceral
video evidence – is seen as epitomizing a modern manifestation of the old maxim attributed
to Clarence Darrow that a case is won or lost by the time the jury is sworn in.3
In this paper, we assess the degree to which such concerns are justified along several dimensions. First, we ask if certain demographic and socioeconomic characteristics are indeed
related to the sentiments and biases that individual jurors take with them into the deliberation room or to the verdicts at which juries collectively arrive. In examining this question,
we do not focus solely on race, but also jointly consider other juror characteristics – such as
sex, age, religiousness, education, and income – that existing literature has largely neglected.
Second, we place the size of the jury composition effects on verdicts in the context of other
important aspects of the trial, especially the strength of the evidence presented in the courtroom. Finally, we assess whether attorneys are aware of the effects of juror characteristics on
pre-deliberation biases and trial outcomes. In addressing these questions, we take advantage
of a uniquely rich dataset on non-capital felony jury trials held in four major U.S. state trial
courts4 during 2000 to 2001 that includes detailed survey information on various aspects of
the case from all trial participants, including the jurors, the judge, and the attorneys.
We find strong evidence of significant relations between individual jurors’ demographic
and socioeconomic characteristics and their pre-deliberation leanings. In particular, we find
1
The Sixth Amendment of the U.S. Constitution states that “The accused shall enjoy the right to a speedy
and public trial, by an impartial jury of the State and district wherein the crime shall have been committed.”
2
For example, Rothwax (1996) suggests that attorneys “can mold a jury in the hope that it will be swayed
by emotion and innuendo, not fact.”
3
In the initial state trial of the officers, the jury was comprised of ten Caucasians, one white Hispanic,
and one Asian. Three of the officers were acquitted of all charges, while the jury was hung on one charge
and acquitted on others for the fourth officer. The infamous videotape of the beating was, of course, not
the only evidence presented at trial, and it remains difficult to predict if the trial outcome would have been
different if a jury of a different racial composition had heard all of the same evidence, testimony, and judge
instructions. See Linder (2007) for further details and discussion.
4
Los Angeles (CA), Maricopa (AZ), Bronx (NY), and Washington D.C.
2
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
that jurors with higher levels of education and religiousness and, most consistently, those
with higher income are significantly more likely to hold biases and sentiments favoring the
prosecution compared to other jurors who heard the same case.5 Female jurors hold some
pre-deliberation opinions weakly favoring the prosecution that depend to some degree on the
gender of the victim. Black jurors show greater favoritism for the defense in general, and
especially tend to express pro-defense sentiments when the defendant is also black.
We then examine whether the juror characteristics that are associated with significant
individual pre-deliberation biases also have strong effects on the verdicts arrived at post
deliberation. Using a rich set of case-level controls, including attorney demographics and
experience measures as well as the judge’s assessment of attorney skills and trial evidence,
we find that juries with higher average income and religiousness hand out significantly fewer
acquittals, closely mirroring our findings regarding jurors’ pre-deliberation biases. We also
find that juries with a greater proportion of blacks convict on fewer counts, predominantly
when both the defendant and victim are black. Finally, we find that a greater proportion
of females and greater average age and years of education amongst seated jurors are weakly
associated with trial outcomes favoring the defendant. These jury composition effects are
small compared to the effect of evidentiary strength, and account for less of the explained
variation in verdicts than do the factual and legal aspects of the trial. Specifically, reducing
acquittals by the same amount as would be induced by increasing the strength of the evidence
by one standard deviation would require, other things equal, replacing four black jurors with
non-black jurors on a twelve-person jury or increasing the income of each juror by about
40%.
Having established the significance of jury characteristics in shaping juror predispositions
and trial outcomes, we examine whether attorneys correctly anticipate these associations by
asking if they hold preferences for juror characteristics that tend to favor their side. Capitalizing on the availability of a self-reported measure of attorney satisfaction regarding the
seated jury, we ask whether the prosecutor’s and the defense attorney’s levels of satisfaction can be explained by the characteristics of the seated jury, once again conditioning our
analysis on a rich set of controls.6 Robust to all specifications, we find that seated juries
with higher average income are associated with significantly higher satisfaction levels for
the prosecution and lower satisfaction levels for the defense. This is fully consistent with
our findings that higher income is strongly related to pro-prosecution juror sentiments and
5
The particular dimensions along which we examine juror bias include sympathy for the victim, trust in
the police, subjective evaluation of each side’s overall case, and interpretation of the trial evidence.
6
We also limit our analysis to cases in which attorneys reported their satisfaction level with the seated
jury before learning the verdict, so that their evaluations should not be influenced by the eventual trial
outcome.
3
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
fewer acquittals. On the other hand, in contrast to the weak association between gender and
verdicts and to the tendency for female jurors to hold some biases favoring the prosecution,
we find strong evidence that defense attorneys prefer having more women on the jury.
Our results regarding prosecuting and defense attorney preferences for blacks on the
seated jury are inconclusive. While the effects of race on attorney satisfaction are sometimes
large, they are often very imprecisely estimated. Further, they suggest that prosecuting
attorneys may sometimes prefer juries with a greater proportion of blacks, in contrast to our
findings that black jurors tend to hold strong predispositions in favor of the defense in general.
We speculate on some factors that may be confounding the estimation and interpretation of
these race effects on attorney satisfaction, including the fact that it is ostensibly illegal to
remove potential jurors on the sole basis of race during jury selection.7
Our analysis improves upon existing studies in several ways. First, while past studies
of the impacts of jury composition on trial outcomes have focused predominantly on race
(e.g. Anwar et al., 2012b; Lee, 2010; Bowers et al., 2001), our results demonstrate that other
demographic and socioeconomic characteristics also play an important role in shaping the
biases of juries. Furthermore, accounting jointly for race and other characteristics correlated
with race allows us to disentangle effects that may have been misattributed solely to race in
past studies that omit these other characteristics.8 Second, our dataset contains trials from
multiple courtrooms across several jurisdictions with varying racial diversities and histories
of racism. As such, results from this paper should be helpful in assessing the nature and the
extent of jury composition effects on trial outcomes, as well as the degree to which attorneys
have preferences over certain juror characteristics, outside the deep southern states and across
counties with varying racial mixes.9 Third, we take a stated preference approach to assessing
7
The Supreme Court ruling in Batson v. Kentucky (1986) and other subsequent rulings forbid attorneys
to remove jurors solely on the basis of race or gender. However, the effectiveness of these rulings has been
widely questioned. This will be discussed in more detail in the following section.
8
Sommers and Ellsworth (2003) review evidence from the legal literature on the effects of race in criminal
jury trials. Also see Pfeifer (1990) for a critical appraisal of some of the earliest of this evidence. Lieberman and Sales (2007) and Baldus et al. (2001) also review evidence on the effects of gender and other
socio-demographic characteristics. The studies reviewed by these authors are often based on small samples
constructed from archival material, on mock trials, or on public opinion polls, and consider the selected
juror characteristics in isolation rather than jointly. In contrast, Anwar et al. (2012b) account for age and
gender jointly with race, while Anwar et al. (2012a) undertake a similar joint analysis focusing on age,
both employing a large dataset on actual trials. Shayo and Zussman (2011), Alesina and La Ferrara (2011),
Abrams et al. (2011), and Iyengar (2011) provide evidence that the race of judges, victims, and defendants
can affect trial outcomes in other settings.
9
Anwar et al. (2012a,b) use data from Lake County and Sarasota County in Florida, which are predominantly white. Rose (1999) uses data from a county in North Carolina described as “largely biracial”. Bowers
et al. (2001) use data from the Capital Jury Project, which encompasses trials from 14 states, though 11 of
these are in the south. Diamond et al. (2009) use data from just a single courtroom, though in the more
racially diverse Cook County, Illinois. Baldus et al. (2001) use data from Philadelphia.
4
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
which juror characteristics are preferred by attorneys on both sides. Past studies have instead
taken a revealed preference approach by exploring the types of potential jurors removed by
each attorney, which may be more susceptible to confounding effects of constraints faced by
attorneys during jury selection.10 Finally, due to the richness of our dataset, we are able
to incorporate into our analysis many aspects of trials and participants that are unobserved
in most previous studies, including, most notably, information on the evidence presented at
trial and jurors’ interpretation of that evidence.
Overall, our results give some justification for concerns about a lack of impartiality and
trustworthiness in U.S. criminal jury trials. One implication is that defendants face the
prospect of an unfair trial simply by virtue of being unlucky in the characteristics of the
jurors drawn from the jury pool on the day of the trial. But an even more disconcerting
implication is that attorneys, with at least some apparent ability to correctly anticipate
relations between juror characteristics and biases, might attempt to use this knowledge to
manipulate trial outcomes. However, our results also provide some important context for
such concerns. First, our multidimensional approach reveals that the sources of jury bias
are more nuanced than an analysis based on race alone would suggest. Second, our results
on attorney preferences indicate that attorneys have only a partial understanding of these
sources of jury bias, which may limit their opportunities to attempt to leverage them to their
advantage. Finally, our comparison of the effects of jury composition with that of evidentiary
strength suggest that the fundamental issues of fact and law remain the primary determinants
of the outcome of any given trial. Investigating these issues further and exploring how our
results generalize to regions and types of cases not covered by our dataset are crucial next
steps for evaluating the fairness of jury trials and the validity of the verdicts delivered by
the U.S. justice system.
The rest of the paper is organized as follows. Section 2 provides a description of the jury
selection process and other relevant institutional details. Section 3 describes the dataset
and the restricted sample of cases used in our empirical analysis. Section 4 examines the
link between juror characteristics and both individual biases and trial verdicts. In Section
5, we present our investigation of attorney preferences over juror characteristics. Finally, we
conclude and discuss some additional policy and research questions that our findings raise.
10
This method has been implemented by Anwar et al. (2012a) to identify attorney preferences over juror
age; by Baldus et al. (2001) to identify attorney preferences over juror race, gender, and age; by Grosso
and O’Brien (2012) and Rose (1999) to identify attorney preferences over juror race; and, for civil trials, by
Diamond et al. (2009) to identify attorney preferences over juror race, gender, age, and income. The jury
selection process and the various restrictions that can be imposed upon attorneys during the process will be
discussed in the following section.
5
Jury Composition, Trial Outcomes, & Attorney Preferences
2
Lehmann & Smith
Institutional Background
Throughout our analysis, we will be using data on the characteristics of seated jurors. It is
thus important to understand the many forces that can influence which jurors come to be
seated on the jury for a given criminal trial.
The broad goal of the overall jury selection process is to randomly draw a panel – also
called a venire – of potential jurors from the population of the county in which the alleged
crime was committed; then, through a procedure known as voir dire, to remove any potential
jurors with an inability to be impartial from the possibility of serving on the seated jury. Of
interest here are the details of how the process is actually implemented, and how this broad
goal can fail to be met.11
There are a number of reasons why the initial panel may not be a random draw from
the county population. Juror rolls, from which the names of those to be summonsed on a
particular day are randomly drawn, are usually constructed from drivers’ license, tax, voter
registration, or other administrative databases, and these can systematically under-represent
certain groups. Summonses that are undeliverable, never responded to, or received by those
with automatic exemptions can likewise over-represent certain groups.12
There is also the question of which cases actually make it to a jury trial in a certain
county. Defense attorneys are occasionally successful at requesting a change of venue into
or out of a given county on the grounds that it would be too difficult to form an impartial
jury in the original trial location. Defendants sometimes request a bench (judge-only) rather
than a jury trial. And the set of cases that reach a conclusion prior to a jury rendering a
verdict or are never prosecuted is determined by a series of decisions and negotiations made
by the court and attorneys on both sides. All of these issues raise the possibility that the
types of cases heard by a jury in a given courtroom can depend to some extent on attorney
expectations of the types of jurors who will eventually comprise the jury.
We will not attempt to diagnose or account for potential non-randomness in the venire or
in the trials represented in our dataset. As with past studies, we have no satisfactory means
to accomplish this. At the same time, we note that the states covered by our dataset have
been above-average or amongst the leaders in jury reform efforts related to jury pool diversity
(Mize et al., 2007), and that defense motions for changes of venue are rarely successful outside
11
Gobert et al. (2009) and Starr and McCormick (2001) each provide a very comprehensive description
of jury selection, from both a theoretical and practical perspective. This section draws primarily from these
two sources and from a number of interviews we have conducted with legal professionals who have experience
with the process.
12
See Mize et al. (2007), Sommers (2008), and American Civil Liberties Union of Northern California
(2010) for further details.
6
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
of very high-profile cases.13 We, therefore, do not expect these issues to substantially affect
our results. On the other hand, voir dire presents a serious potential source of endogeneity,
which we make a concerted effort to redress in our analysis. Voir dire can be leveraged
by attorneys for far more than simply identifying potential jurors with an inability to be
impartial. In fact, lawyers are taught to identify potential jurors who are likely to hold
unfavorable predispositions to their side and to attempt to remove them. This often takes
the form of targeting specific socio-demographic groups that community surveys or mock
trials have suggested are more likely than others to hold certain opinions on various aspects
of the case.14 However, attorneys face a number of institutional constraints in this endeavor
and must also consider the strategies of their opposing counsel.
During voir dire, potential jurors are questioned, and on the basis of their responses, can
be deselected from (rather than selected to) the seated jury. The sole legal rationale for voir
dire is to ensure that the seated jury is impartial. Individual judges have wide latitude in
deciding how this is to be accomplished: there are no constitutionally-mandated procedures,
and while most states and some local courts have guidelines or a set of common practices in
place, these are rarely binding for any given trial.
The examination of potential jurors can be carried out primarily by the judge, by the
judge with suggestions or a greater degree of participation from the attorneys, or primarily
by the attorneys. It can also be carried out in the open courtroom or, less often, privately
in the judge’s chambers or at the sidebar. The primary concerns that are examined in any
voir dire are the existence of personal relationships between panelists and any other parties
involved in the trial, and the capacity to understand and contemplate the relevant legal
issues and evidence in a dispassionate and impartial manner. But many other subjects are
often explored as well, especially if attorneys have a high degree of participation. Attorneys
may also have access to the results of questionnaires filled in by potential jurors: in almost
all cases, they observe the information on the potential jurors that the court collects, which
includes basic demographic and occupational data at the least; and they are sometimes
also permitted to design and administer “Supplementary Juror Questionnaires” (SJQs) to
elicit more detailed information. The judge has discretion over each of these facets of the
examination phase; individual judges tend to have strong views on how these decisions should
13
For example, litigation consultant Gary Moran has stated that such requests are often motivated primarily by delay and diversion. He concludes: “The judge is going to sit there and sit there and after four
hours . . . he will finally utter the magic word – ‘Denied.’ And then we all move along” (quoted in Kressel
and Kressel, 2002, 59).
14
In addition to Gobert et al. (2009) and Starr and McCormick (2001), which include much practical
instruction for attorneys, see also Hoffman (2006) and Kressel and Kressel (2002) on attorney strategies,
and Lieberman and Sales (2007) and Posey and Wrightsman (2005) on the professional practice of what has
come to be known as scientific jury selection.
7
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
be made, and to hold adamantly to them.
Attorneys can remove panelists via two instruments: strikes for cause and peremptory
strikes. An attorney on a given side can challenge a potential juror for cause by arguing that
an inability to be impartial has been demonstrated, and the challenge can then be debated
amongst the attorneys and the judge. If the judge rules in favor of the strike, the potential
juror is dismissed. Attorneys can strike an unlimited number of potential jurors for cause,
as long as the judge can be sufficiently convinced that a basis for disqualification has been
demonstrated. Each attorney can also remove a limited number of potential jurors without
stating any reason by exercising peremptory strikes. The total number of peremptory strikes
available is generally small relative to the size of the panel and is usually divided equally
between the defense and prosecution, though again, these matters are ultimately at the
discretion of the judge.15
Although attorneys ostensibly have full discretion over how to exercise their peremptory
strikes, they are in fact legally obliged to do so in a non-discriminatory manner. If an
attorney suspects that the opposing counsel has exercised a peremptory strike solely on the
basis of a potential juror’s race or gender, a “Batson objection” can be raised.16 When a
Batson objection is raised and sustained by the judge, the opposing counsel must provide a
race- or gender-neutral reason for the strike in order for it to be allowed. If such a reason is
not provided, the strike will be disallowed, and other remedies can occasionally be imposed
at the judge’s discretion as well, even leading to a new voir dire or a mistrial in extreme
cases. Many legal commentators regard Batson as only a minor impediment to how attorneys
exercise peremptory strikes, since attorneys can and often do comply with its requirements by
stating neutral but extremely arbitrary and improbable reasons for the strikes.17 Less direct
ways in which Batson might influence attorney behavior, such as heightening concerns over
reputation or the likelihood of verdicts being overturned on appeal, have not been studied.
This description indicates that attorneys have some control over the composition of juries, with the degree of control increasing in the number of peremptory strikes available and
15
The order in which strikes are made – across attorneys and types of strikes – depends to some extent on
whether the struck jury method or some version of the strike-and-replace method is being employed. These
are futher details that can vary widely by trial and courtroom, though they are less relevant to the present
discussion. See Gobert et al. (2009) and the related discussion in Lehmann and Smith (2012) for more on
variations in the sequence in which voir dire can be conducted.
16
The seminal ruling in Batson v. Kentucky (1986) established that the prosecuting attorney must be able
to state a race-neutral justification if the defense counsel can make a prima facie case that a peremptory
strike has been exercised against an African American potential juror solely on the basis of race. Further
Supreme Court rulings have extended the applicability of Batson to peremptory strikes by defense as well
as prosecuting attorneys, to all races, to discrimination on the basis of gender as well as race, and to civil as
well as criminal trials.
17
See, for example, Equal Justice Initiative (2010) and Norton et al. (2007).
8
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
in the freedoms afforded to attorneys in eliciting information about the potential jurors.18
Attorneys can thus influence trial outcomes through multiple channels: by affecting which
jurors from the pool of potential jurors with heterogeneous biases will hear the case; and
through the usual processes of argument and presentation of evidence during the trial itself.
One would thus not be surprised to find correlations between verdicts and jury composition,
since both can be affected by common elements of attorney strategies, which themselves can
depend on other aspects of the case. In order to isolate the causal effect of jury composition
on trial outcomes, we will thus need to address this endogeneity. We intend to do so by controlling for a rich set of covariates reflecting the freedoms and constraints faced by attorneys
in voir dire, the attorneys’ skills in designing and implementing effective strategies, and the
aspects of each trial that attorneys might condition their strategies on. We will return to this
in Section 4, but will first introduce our dataset and the wealth of information it contains.
3
3.1
Data Description
The NCSC/ICPSR Hung Juries Dataset
Our empirical analysis relies on a detailed examination of 351 felony jury trials in four
major county courts across the U.S. collected by Hannaford-Agor et al. (2002, 2003) for
the National Center for State Courts (NCSC) during 2000 and 2001 and disseminated by
the Interuniversity Consortium for Political and Social Research (ICPSR). The four courts
– Los Angeles County Superior Court in California, Maricopa County Superior Court in
Arizona, Bronx County Supreme Court in New York, and District of Columbia Superior
Court in Washington, D.C. – were chosen based on their high volume of felony jury trials
and their willingness to cooperate with the data collection process and guidelines. Although
the NCSC’s main goal in the study was to provide an empirical evaluation of hung juries,
the dataset is not limited to trials ending in a hung jury, and indeed includes all non-capital
felony trials held in these courts during the specified periods of data collection.19
The NCSC study provides a comprehensive look at the trial and all the parties involved:
the defendant(s) and the victim(s) (if any), the judge, the attorneys, and, most importantly
18
We provide some theoretical and empirical support for this intuition in Lehmann and Smith (2012), and
also confirm the importance of relative attorney skill. See also Neilson and Winter (2000) and Ford (2010).
19
Each county had a single period of data collection ranging from 4 to 11 months in length, with data
collection ending in Los Angeles in October 2000 before beginning elsewhere, though with the other three
periods overlapping in various months in 2001. D.C. contributes the largest number of trials to the dataset,
though each court is fairly evenly represented. The death penalty was available in all counties except for
D.C. at the time of data collection, though the distribution of charges in the dataset is roughly the same
in each county. It should be noted that the confidentiality rules agreed to by the courts precluded any trial
participant from accessing any of the survey information at the time of the respective trial or afterwards.
9
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
for our analysis, the seated jurors. We do not have information about the initial jury pool
from which the seated jury was formed, except for the initial panel size.20
The case data in the NCSC study provide researchers with an extensive look at all the
charges, the race and sex of the defendant(s) and victim(s), and the voir dire process that led
to the seated jury. The NCSC questionnaires also asked the presiding judge of each trial for
his/her evaluation of the evidence, case complexities, and attorney skills. The main variables
of interest for our empirical analysis come from the attorney and the jury questionnaires.
The attorneys in each trial were asked about their satisfaction with the voir dire process,
legal experience, and basic demographic information, including sex and age. In addition,
the juror questionnaire provides a rich set of demographic information about each seated
juror, including age, sex, race, education, income, and religious beliefs. The NCSC data
also contain a description of the dynamics of each juror’s opinion formation, and of pre- and
post-deliberation perceptions and opinions regarding the defendant(s) and victim(s). To our
knowledge, the richness of the jury demographic and opinion information available in the
NCSC study is unmatched by any other readily available dataset.
Regrettably, although the NCSC dataset contains information on 351 cases and 3,497
jurors in total, not every juror in our data answered every survey question. Since much of
our analysis relies on the average characteristics of the seated juries, we restrict our sample
of interest to cases in which these average characteristics are well-measured. Specifically, we
focus on trials for which six or more jurors responded to questions about education, income,
age, race, gender, and religion to reduce any bias that may result in our calculation of average jury characteristics due to systematic non-reporting.21 Additionally, we further restrict
our sample to those cases in which attorneys reported their voir dire satisfaction measures
before learning the verdict, in order to minimize the possibility that our measure of voir
dire satisfaction is confounded by satisfaction with broader aspects of the trial, specifically
outcomes.
20
We can, however, infer the average characteristics of the panel from those of the county as a whole, since,
as was discussed previously, a panel is essentially a random sample of the county population. We include
county fixed effects in all of our empirical specifications, in part to account for the composition of the panel
as much as possible.
21
Small changes to this cut-off point do not alter out main findings. All counties represented in our dataset
were required to have twelve-person juries for felony trials at the time of data collection, except Maricopa,
where eight-person juries were permitted for felony trials with charges below a certain level of severity. We
do not directly observe which trials in Maricopa actually had eight-person juries, but using a smaller cut-off
for all trials that were eligible for eight-person juries does not affect our results.
10
Jury Composition, Trial Outcomes, & Attorney Preferences
3.2
Lehmann & Smith
Summary Statistics
Table 1 provides some descriptive statistics of our restricted sample. Panel A shows that the
majority of cases in our restricted sample fall into the categories of property or drug-related
crimes. Consequently, a possible criticism of our dataset is that most of our cases are restricted to non-capital felonies that are not as “serious” as homicides (actual or attempted) or
sexual crimes. In turn, attorneys might not pursue the same strategies, and jurors might not
act as strongly on their personal prejudices as would be the case with more serious charges.
Therefore, the results that we find might be underestimates of the level of bias present in
more serious cases. While a valid concern, we believe that our results remain interesting
for the following reasons. First, many of the previous studies on bias and discrimination
in jury trials have focused on capital crimes. However, we believe that it is important to
assess potential impediments to fairness across various types of trials/crimes that are much
more prevalent and frequent in courtrooms. Second, in all of our regression specifications, we
control for the type of crime. Moreover, in alternative specifications, we interact our main
variables of interest with an indicator of whether the crime is a murder or a sex crime and
find no significant differences in our main results across case types.
Panel B shows the case-level means of the various sets of controls on which we rely in
our empirical analysis. The case descriptive variables show that the majority of the cases
have black defendants and black victims. Over half of the cases are represented by a public
defender, which generally speaks to the low income levels of the defendants. There is wide
variation in the voir dire process, from who conducted the voir dire22 to the length of time
it took to shape the final jury. A Batson objection was raised in about 14% of the cases. A
typical case in our sample involves close to three different charges against the defendant(s).
Of these charges, about 56% result in a conviction, 38% in an acquittal, and the remaining
6% in a hung jury.
The next set of variables summarizes the average characteristics of the seated jury. These
are our main variables of interest. In the jury supplemental survey, jurors only provide
categorical responses to questions about their education, income, and age. In order to
facilitate a more useful interpretation of our coefficients, we translate the categorial variables
into appropriate levels for each variable by taking the mid-point of each category. However,
using the categorical variable directly does not change our main results. In our restricted
sample, a typical seated jury has jurors with about 15 years of education who earn $58,000 in
annual income and are about 42 years old. On average, roughly 66% of the jurors seated for
22
“Who conducted” is a variable ranging from 1 to 4, increasing in attorney involvement, where 1 is “Judge
with little or no attorney involvement” and 4 is “Attorney with little or no judge involvement.”
11
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 1: Case Summary Statistics
Panel A:
% of Cases
16.18
2.94
39.71
41.18
N
22
4
54
56
Mean
Std. Dev.
N
0.594
0.231
0.333
0.098
0.555
0.519
2.713
1.415
0.493
0.423
0.473
0.299
0.499
0.502
2.277
1.975
138
134
135
132
137
133
136
135
2.309
0.223
5.343
0.273
0.137
0.923
0.418
6.278
0.447
0.345
136
139
137
139
139
15.101
57.603
41.523
0.663
0.569
0.261
0.508
0.248
1.090
12.462
3.982
0.160
0.179
0.235
0.502
0.434
139
139
139
139
139
139
130
133
Defense:
Voir Dire Satisfactory (1 “Low” to 7 “High”)
Legal Practice (years)
Previous Criminal Trials
Age
Female
5.403
12.903
62.323
41.880
0.313
1.483
8.459
73.089
10.239
0.465
139
134
133
133
131
Prosecution:
Voir Dire Satisfactory
Legal Practice (years)
Previous Criminal Trials
Age
Female
5.396
8.696
39.584
36.372
0.495
1.377
6.569
57.515
7.913
0.502
111
112
113
113
105
Case Type
Murder (1st, 2nd, attempted) and Manslaughter
Sexual crime
Robbery, Burglary, Larceny, Theft, Assault, Arson
Others, including drug-related crimes
Panel B:
Variable
Case Characteristics:
Victim
Female Victim
Non-black Victim
Female Defendant
Black Defendant
Public Defense
Total Number of Charges
Number of Charges Convicted
Voir Dire Characteristics:
Who Conducted (1 “Judge” to 4 “Attorney”)
Questionnaire Used
Voir Dire Time (hours)
Anonymous Jury
Batson Objection Raised
Jury Characteristics:
Education (years)
Income ($, thousands)
Age
Religious
Female
Black
Woman Foreperson
Black Foreperson
Attorney Characteristics:
Notes: Sample limited to cases with more than 5 jurors responding to school, income, age, race, gender,
and religion questions, and non-missing site information. Also restricted to cases in which both the defense
and the prosecution answered voir dire satisfaction question before learning verdict.
12
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
a particular case describe themselves as religious.23 A typical jury in our restricted sample is
over 50% women and about a quarter black, and the sex and race of the foreperson generally
reflect the proportions in the overall jury population.24
Throughout our empirical analysis, we estimate specifications in which we control for attorney characteristics, because the attorney’s satisfaction with the voir dire is a self-reported
variable and his/her experience or sex may influence his/her evaluation of the seated jury.
The mean level of satisfaction is the same for the defense as for the prosecution, and the
dispersion of the satisfaction measure is similar for the two sides as well.25 A typical defense
attorney in our sample is older, has much more legal experience, and is more likely to be
male compared to a typical prosecutor. Unfortunately, our data do not include the race of
the attorneys.
4
Juror Characteristics and Biases
In this section, we first explore the link between individual juror characteristics and the
pre-deliberation opinions held by jurors. Next, we examine the role of jury composition in
shaping trial outcomes by assessing the relationship between average jury characteristics and
verdicts. We then put these jury composition effects on verdicts in the context of the effect
of evidentiary strength, treating the judge’s assessment of the trial evidence as an objective
evaluation of the case against the defendant.
4.1
Pre-Deliberative Juror Biases
The survey of jurors in our dataset asks a series of questions about opinions concerning the
defendant, the victim, the attorneys, and other aspects of the case, as well as crime more
generally. We ask whether differences in these stated opinions across jurors within each
case can be explained by the jurors’ demographic and socioeconomic characteristics. More
23
Answered 1 or 2 on a scale of 1 to 5 where 1 is very religious, 2 is religious, and 5 is very nonreligious.
Comparing our restricted sample with the county population demographics, we find that females are
overrepresented in our sample in Bronx (62.1% in NCSC sample versus 53.4% in the Census) and D.C. (64.3%
in the NCSC sample versus 52.9% in the Census) while blacks are overrepresented in L.A. (15.8% versus
9.4%) and Bronx (40.4% versus 31.9%) and underrepresented in D.C. (44.9% versus 59.1%) The median
household income of the seated jurors in our sample are significantly higher than in the county population
estimates from the Census by about $15,000 to $20,000. These differences can be partially attributed
to the selective pools from which initial jury summonses are drawn. County demographic estimates are
from U.S. Census Bureau Population Estimates, April 1, 2000 to July 1, 2010. Retrieved March 8, 2011
from http://www.census.gov/popest/estimates.html. Household median income data from U.S. Census
Bureau Small Area Income and Poverty Estimates. Retrieved March 15, 2011 from http://www.census.
gov/did/www/saipe/data/index.html.
25
We will discuss this variable in more detail in Section 5.
24
13
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
specifically, we estimate the case fixed effects model
yjc = βXjc + αc + εjc ,
(1)
where yjc is the level of agreement of juror j with various statements about case c and the
parties involved (on a scale ranging from 1 “Strongly Disagree” to 7 “Strongly Agree”), Xjc is a
vector of juror characteristics and selected interactions with some case characteristics, and αc
captures case-level characteristics.26 Introducing case fixed effects ensures that variations in
jurors’ opinions cannot be attributed to differences in unobservable case-level characteristics.
Table 2 reports OLS estimates of equation (1) for six separate questions that jurors were
asked to state an opinion on. While this linear modeling approach ignores the ordinal nature
of the dependent variables, alternative models that would account for this ordinality, such
as the ordered logit or ordered probit, encounter the incidental parameters problem within a
fixed effects framework, which renders the usual maximum likelihood estimators inconsistent
even when the models are properly specified (Greene and Hensher, 2010). We have chosen
to focus on the results from the linear fixed effects models to ensure that our estimates are
reliable and free from the confounding effects of unobserved within-case factors.27
Table 2 provides evidence of significant relations between the education, income, and
religiousness of individual jurors – characteristics that have been largely ignored in past
studies on jury discrimination – and their pre-deliberation leanings. Compared to others who
heard the same case, jurors with higher education levels are less likely to express sympathy
for the victim (column 1). More educated jurors are also significantly more likely to interpret
the trial evidence as strongly favoring the prosecution relative to the defense (column 6).
Specifically, an extra year of education is associated with a 0.03 point decrease on the 7-point
scale that measures juror interpretation of evidentiary strength.28
Most consistently across the six predispositions, jurors with higher income hold predeliberation leanings that are more favorable for the prosecution. While individuals with
26
We cannot observe whether the same jurors appear in more than one case. However, given the short
period of data collection in each county, it is highly unlikely that one individual would have been summonsed
for jury duty more than once within that period, let alone making it on to the seated jury more than once.
We therefore treat each observation in our dataset as representing a distinct juror.
27
A solution that allows both unobserved heterogeneity and ordinality to be accounted for is presented
by Baetschmann et al. (2011), who propose and demonstrate the consistency and favorable small-sample
properties of an alternative fixed effects ordered logit estimator. Utilizing this BUC estimator produces
estimates that are qualitatively similar though less precise than those from our linear models. Note that
each specification in Table 2 is estimated independently: attempts to account for cross-equation correlations
also produced less precise but substantively similar estimates.
28
The dependent variable in column 6 is the juror’s response to the question “All things considered, how
close was the case?” to which the juror can answer on a scale of 1 to 7 where 1 is “Evidence strongly favored
the prosecution” and 7 is “Evidence strongly favored the defense.”
14
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
higher earnings are more critical of attorney skills for both the prosecution and the defense,
we find that higher income is significantly associated with less sympathy for the victim
(column 1), greater trust in police in the community (column 2), and a greater likelihood of
evaluating the trial evidence in favor of the prosecution (column 6). Specifically, a $10,000
increase in juror income has the same effect as an additional year of education (i.e. a 0.03
point decrease on the 7-point scale) on the interpretation of the trial evidence. Similarly,
jurors who self-identify as “Religious” or “Very Religious” express significantly greater trust in
the police, a higher rating of the prosecution’s case (column 4), and an interpretation of the
evidence inclined towards the prosecution. Thus, higher education, income, and religiousness
appear to be associated with a general bias towards the prosecution.
Table 2 also reveals important sex and race effects on individual juror biases before group
discussion. For clarity and ease of interpretation, the middle panel in Table 2 reports the
full sex and race effects under different sex/race composition identities of the victim and
defendants in the case.29 Female jurors tend to rate the skill of the prosecuting attorney
more highly (column 3) in all cases, but appear to have some other biases that depend to
some degree on the gender of the victim. For example, female jurors are weakly less likely
than other jurors to report sympathy for the victim or a high rating of the defense’s overall
case when the victim is male, but display a smaller difference in these regards compared to
other jurors when the victim is female. However, the coefficents are imprecisely estimated,
and equality of the gender effects by sex of the victim cannot be ruled out. While firm
conclusions are more difficult to draw for the case of juror gender, the greater tendency
to rate the prosecutor’s skill highly and to regard the defense’s case as weak indicate that
women may have a moderate general bias towards the prosecution.
Examining the relation between juror race and pre-deliberative biases, we find large effects consistently across the six predispositions and most combinations of defendant and
victim race. Black jurors report significantly lower trust in the police, lower ratings of the
prosecutor’s skill and case, higher ratings of the defense’s case, and an interpretation of the
evidence more in favor of the defense. The divergence in evaluations of the prosecutor’s
case and the defense’s case by black jurors relative to other jurors is most prominent when
the defendant is black. On the other hand, the divergence in the evaluation of the prosecutor’s skill by black jurors relative to other jurors is most prominent when the defendant
is not black. Finally, although the race effect on the interpretation of the evidence is only
statistically significant when both the defendant and the victim are black, the magnitudes
are similar for cases involving all defendant/victim race combinations. Specifically, black
29
For example, the full black effect when the defendant and the victim are black (i.e. dblack = 1 and
vnblack = 0) is the sum of coefficients on black and black × dblack.
15
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 2: Pre-Deliberation Juror Opinions with Case Fixed Effects: OLS Results
Dep. Var. = 1 to 7, where 1 is “Strongly Disagree” and 7 is “Strongly Agree”
education (years)
income ($, thousands)
age
religious
female
female × vfemale
black
black × dblack
black × vnblack
Female effect when
vfemale = 0
vfemale = 1
Black effect when
dblack = 0, vnblack = 0
dblack = 1, vnblack = 0
dblack = 0, vnblack = 1
dblack = 1, vnblack = 1
N
# of Cases
F-stat
Sympathy
for Victim
Trust
Police
(1)
(2)
-0.113**
-0.001
(0.053)
(0.018)
-0.013*** 0.005***
(0.005)
(0.002)
-0.000
0.002
(0.010)
(0.004)
0.027
0.251***
(0.225)
(0.084)
-0.560
0.076
(0.443)
(0.153)
0.626
-0.064
(0.405)
(0.198)
0.593
-0.434*
(0.655)
(0.256)
-0.155
0.051
(0.680)
(0.276)
-0.272
-0.336
(0.666)
(0.300)
Skillful Prosecution
Defense
Prosecutor Case Strong Case Strong
(3)
(4)
(5)
How
Close
(P → D)
(6)
0.008
(0.020)
-0.004**
(0.002)
0.003
(0.003)
0.050
(0.090)
0.307**
(0.133)
0.063
(0.175)
-0.480**
(0.188)
0.342
(0.221)
0.051
(0.245)
0.016
(0.016)
-0.000
(0.002)
-0.001
(0.003)
0.211***
(0.081)
0.066
(0.143)
-0.218
(0.202)
-0.079
(0.207)
-0.243
(0.226)
-0.126
(0.235)
-0.008
(0.017)
-0.004***
(0.002)
-0.003
(0.003)
0.027
(0.081)
-0.319*
(0.162)
0.225
(0.165)
0.032
(0.236)
0.194
(0.254)
0.350
(0.290)
-0.031*
(0.016)
-0.003*
(0.002)
-0.002
(0.003)
-0.148*
(0.082)
-0.089
(0.162)
0.150
(0.183)
0.380
(0.234)
-0.029
(0.261)
-0.109
(0.286)
-0.560
(0.443)
0.066
(0.404)
0.076
(0.153)
0.012
(0.204)
0.307**
(0.133)
0.371**
(0.190)
0.066
(0.143)
-0.153
(0.185)
-0.319*
(0.162)
-0.094
(0.154)
-0.089
(0.162)
0.062
(0.177)
0.593
(0.655)
0.438
(0.379)
0.320
(0.555)
0.166
(0.664)
1418
169
1.693
-0.434*
(0.256)
-0.383**
(0.162)
-0.770***
(0.212)
-0.719**
(0.312)
1406
169
3.603
-0.480**
(0.188)
-0.138
(0.160)
-0.429*
(0.237)
-0.087
(0.254)
1400
169
1.296
-0.079
(0.207)
-0.322**
(0.141)
-0.206
(0.225)
-0.449**
(2.227)
1409
169
1.245
0.032
(0.236)
0.227*
(0.132)
0.382*
(0.236)
0.577*
(0.295)
1403
169
2.420
0.380
(0.234)
0.351**
(0.147)
0.271
(0.233)
0.241
(0.292)
1411
169
2.102
Notes: Robust standard errors clustered at the case level are reported in parentheses. Sample limited to
those cases with more than 5 jurors responding to school, income, age, race, gender, and and religious
questions and to those cases in which both the defense attorney and the prosecutor answered the voir dire
satisfaction question before learning verdict. “How close” asks “All things considered, how close was the
case?” and the juror can answer from 1 to 7 where 1 is “Evidence strongly favored prosecution” and 7
“Evidence strongly favored defense.” All regressions also control for the juror’s employment status, previous
jury experience, previous criminal trial experience, and his/her race and sex interacted with the race and
sex of the foreperson. ∗=10%, ∗∗=5%, ∗ ∗ ∗=1%.
16
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
jurors report an interpretation of the evidence that is rougly 0.25 to 0.35 points higher on
the 7-point scale than other jurors. To put the magnitude of this race effect in context, an
equivalent change in the interpretation of evidence would require a drop in income of about
$100,000 or a decline in education of about 10 years. Thus, black jurors appear to have a
strong general bias towards the defense.
4.2
Average Jury Characteristics and Verdicts
Having demonstrated that there are significant relations between individual juror characteristics and pre-deliberation leanings, we now explore whether these individual biases feed
through into biases in the determinations of guilt that juries make collectively. More precisely, we will assess the relation between the average characteristics of the seated jury and
the trial verdicts. Before proceeding with the analysis, we first address some methodological
issues.
While our dataset includes information on how individual jurors voted at various stages of
the deliberations, we have chosen to focus on actual trial verdicts rather than individual juror
voting patterns. This choice was motivated by several considerations. First, as noted by some
legal scholars, the human dynamics involved in jury deliberation provide a potential channel
through which individual prejudices can be attenuated or amplified within a group-decision
setting.30 We therefore consider the question of how jury composition relates to the verdicts
arrived at when deliberations have concluded to be of much more policy relevance than the
question of how individual juror characteristics relate to individual juror votes at earlier
stages of deliberation or to how individual jurors would have decided the case on their own.
Second, while estimating a case fixed effects model relating the voting patterns of individual
jurors to their characteristics might mitigate bias due to unobserved heterogeneity, it would
necessarily entail dropping all cases in which all jurors voted the same way. Since hung
juries occur for only 8 percent of the counts in our estimation sample, this restriction would
drastically reduce our sample size and statistical power. Finally, our dataset only contains
information on individual jurors’ votes on the “most serious charge” against the defendant.
The wording of the survey questions appears to have led to differing interpretations across
respondents and, more generally, a low response rate, which limits our confidence in the
quality of this information.
30
This idea is related to the notion of “group polarization” found in the psychology literature, which states
that, if a group is like-minded, discussion strengthens its prevailing opinions. In an influential study, Myers
and Bishop (1970) found that talking about racial issues increased prejudice in a high-prejudice group of high
school students and decreased it in a low-prejudice group. Bringing this idea into the legal field, Sunstein
(2000) writes that, in small deliberative groups such as juries, there may be a “robust pattern” of polarization
whereby the initial inclinations of individuals before deliberation become more severe during deliberation.
17
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Having thus decided to focus on verdicts rather than voting patterns, we must still
confront the issue of how to precisely define and specify our dependent variable. Past studies
– most notably Anwar et al. (2012a,b) – have focused on a binary dependent variable
indicating whether there was a conviction on any count for a given trial. However, these
studies report that few of the trials being examined had multiple charges. In contrast, Table
1 shows that the average number of counts per trial in our dataset is 2.7, with a wide variance
around this mean. We consider the loss of information that would result from the collapse
of the outcomes on such a large number of charges into a binary conviction indicator to be
undesirable. The alternative that we have adopted is to represent trial outcomes with a pair
of count variables, indicating the absolute number of convictions and acquittals for a given
trial respectively. While we do not attempt to account for varying seriousness of charges or
to otherwise weight individual counts, our approach allows us to distinguish outcomes more
finely – and perhaps also more in line with the objectives of attorneys on each side – than
an approach based on a binary indicator.31 We model trial outcomes thus defined by way of
an independent poisson regression for each dependent variable.32
Finally, we must return to the endogeneity concerns raised by the discussion of the voir
dire process in Section 2. The unit of observation for our analysis of the impact of jury
composition on trial outcomes will necessarily be individual cases, so that it will no longer
be possible to employ case fixed effects.33 This raises the possibility that our results will be
31
Of the 2.7 counts per trial, 1.4 result in a conviction on average, while 1.1 result in an acquittal, and
the remaining 0.2 result in a hung jury. In contrast, headline conviction rates for American criminal trials in
general are typically reported as two thirds or higher (see, for example, the Sourcebook of Criminal Justice
Statistics Online, Table 5.57). However, such headline conviction rates refer to convictions on any charge
for a given trial, including secondary counts and pleas to lesser charges. The relatively low conviction rate
suggested by the average number of acquittals and hung juries per count in our data is precisely the result
of our decision to examine counts within a case individually rather than compressing this information into
a simple any conviction/no conviction assessment for each trial.
32
Our modeling choice has primarily been driven by efficiency concerns. The results that we present below
remain qualitatively similar if we instead use a binary dependent variable, either estimated by way of a
linear probability model as by Anwar et al. (2012b) or a logistic model; and are likewise very similar to
those that are obtained from a linear or negative binomial model applied to the count dependent variables.
However, the results from the poisson model are estimated the most precisely, and are therefore the results
that we focus on. This decision also receives some support from various specification tests, in which the null
hypothesis that the poisson model is the correct one cannot be rejected in our data. While the negative
binomial model would allow us to relax the assumption implicit in the poisson model that the mean and
variance of the dependent variable are the same (which, in any case, is not obviously false in our data),
we instead address this potential concern by using robust standard errors within the poisson model, which
has the additional attractive feature of being robust to misspecification. Unfortunately, our relatively small
sample size also raises precision concerns when attempting to account for cross-equation correlation in our
estimation, for example by way of the seemingly-unrelated poisson approach suggested by King (1989).
33
Our discussion above on the number of charges per trial suggests that we could alternatively treat
individual counts as the unit of observation, which would permit us to control for case fixed effects. However,
since all of our explanatory variables – including, most notably, average jury characteristics – vary at the
trial level only, this approach would not allow us to address our hypotheses of interest.
18
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
biased by unobservable factors related to both verdicts and jury composition. For example,
attorneys might choose jury selection strategies based on observable juror characteristics
jointly with strategies for other aspects of the trial, thereby inducing some simultaneity in
the determination of jury composition and verdicts. We address this endogeneity issue by
controlling for factors affecting attorney strategy and as many other aspects of the trial as
possible. Our dataset contains a rich set of such proxy variables for many potential sources
of unobserved heterogeneity.
Specifically, we control for all of the case, voir dire, and attorney characteristics that
are summarized in Table 1.34 The attorney experience variables, in conjunction with the
voir dire characteristics, are meant to capture both the means and the ability of attorneys to
design and implement jury selection strategies that could affect the composition of the seated
jury. The remaining case-level variables are meant to capture both the direct effects of these
and other characteristics with which they are correlated, as well as any indirect effects that
may result from attorneys attempting to condition their strategies on these variables.
In addition, we include a set of variables describing the trial judge’s opinion of various
aspects of the case, including strength of evidence, attorney performance for each side,
and the complexity of the evidence and the law. Table 3 summarizes the seven variables
accounting for these judge opinions, which we treat as objective assessments of these aspects
of the trials.35 The attorney performance variables provide additional indicators of attorney
ability to shape jury composition and trial outcomes, while the other aspects, in addition to
their direct interest, may also guide or constrain attorney strategies. Moreover, the evidence
variables provide us with a unique opportunity to assess the relative prominence of jury
composition effects as drivers of trial outcomes.36
Table 4 reports our poisson results for the number of convictions and the number of
acquittals. To avoid over-fitting our data and to preserve as many observations as possible,
we include some specifications that limit the control variables included. Columns 1 and 4
and columns 2 and 5 control for measures of the presiding judge’s opinion of the case and
attorney characteristics, respectively, while columns 3 and 6 include both sets of judge and
attorney controls. Moving from left to right, the number of trials for which data on all
34
The voir dire satisfaction variables are excluded. We treat voir dire satisfaction as a separate outcome
variable, and analyze its determinants in detail in the following section.
35
We do not observe identifying or demographic information about the trial judges. It is therefore possible,
in the absence of other proxies for judge effects, that these judge assessments – as well as the voir dire
characteristics, since, as discussed above, judges have a high degree of discretion over the voir dire process –
will also capture the effects of otherwise unobservable judge characteristics. Given the size of the jurisdictions
covered by our dataset, we regard the number of trials presided over by a single judge to be small, though
we ultimately have no way to verify this.
36
Though measured on a seven-point Likert scale, we include these variables linearly in our main specifications. Alternative categorical specifications do not alter our results.
19
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 3: Summary of Judge Opinions
Variable
Evidence presented at trial complex
(1 = “Not at all” to 7 = “Very complex”)
Mean
2.328
Std. Dev.
1.526
N
134
How complex was the law
(1 = “Not at all” to 7 = “Very complex”)
2.597
1.576
134
Evidence favored which side
(1 = “Prosecution strongly” to 7 = “Defense strongly”)
3.120
1.360
133
Attorneys presented all relevant evidence
(1 = “Completely disagree” to 7 = “Completely agree”)
5.561
1.540
132
Skillful prosecutor during trial
(1 = “Not at all” to 7 = “Very skillful”)
5.000
1.535
136
Skillful defense attorney during trial
(1 = “Not at all” to 7 = “Very skillful”)
5.199
1.444
136
How important was police testimony
(1 = “Not at all” to 7 = “Very important”)
5.052
1.669
134
Notes: Sample limited to cases with more than 5 jurors responding to school, income, age, race, gender,
and religious questions. Sample also limited to those cases in which both attorneys answered voir dire
satisfaction question before learning verdict.
control variables in a given specification are available decreases. Restricting the sample to
only those cases included in the specifications in columns 3 and 6 does not change the results
in the other columns.37
Our results reveal that some of the pre-deliberation biases we found to be associated with
various juror characteristics do indeed appear to feed through into the verdicts that juries
with concentrations of these juror characteristics arrive at. Juries made up of higher income
individuals are strongly associated with a decreased number of acquittals. In the specification
controlling for both attorney characteristics and judge opinions (column 6), a $10,000 increase
in the jury’s average income is approximately associated with a 64 percent decrease in the
37
Many control variables are suppressed in Table 4 to save space, but are included in the regressions.
In addition to the variables already discussed, all equations also include controls for county, case type,
and number of counts. Constraining the coefficient on the natural log of the total number of counts in a
given trial to be one would impose the restriction that the rate of conviction or acquittal stays constant
as the number of counts rises. Imposing this restriction makes little difference to the rest of our results,
though the null hypothesis that it holds can be rejected at conventional significance levels. Specifically, the
unconstrained coefficient estimate on the natural log of counts is greater than one in the acquittals equations
and, correspondingly, less than one in the convictions equations, indicating that the rate of acquittal is
actually increasing in the total number of counts. This may indicate that prosecutors add less serious
charges when the main charge lacks evidence in an attempt to secure at least one conviction, or that defense
attorneys focus their efforts on a large number of secondary charges when conviction on the main charge
seems assured; either would lend further support to our assertion that a binary conviction indicator would
seem to miss some potentially important information.
20
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 4: The Effect of Average Jury Characteristics on Verdict: Poisson Results
Dep. Var. = Number of counts that resulted in
Convictions
education (years)
income ($, thousands)
age
religious
female
female × vfemale
black
black × dblack
black × vnblack
Judge - Side Evidence Favored
(1 “Pros.” to 7 “Def.”)
Female effect when
vfemale = 0
vfemale = 1
Black effect when
dblack = 0, vnblack = 0
dblack = 1, vnblack = 0
dblack = 0, vnblack = 1
dblack = 1, vnblack = 1
N
Log-Likelihood
Judge’s Opinion?
Attorney Controls?
Acquittals
(1)
-0.187*
(0.100)
0.010
(0.009)
-0.021
(0.017)
0.446
(0.447)
-0.241
(0.624)
-0.979
(1.205)
-1.029
(1.079)
-0.330
(0.912)
1.507*
(0.798)
-0.297***
(0.065)
(2)
-0.172
(0.158)
0.002
(0.015)
-0.064*
(0.034)
1.036*
(0.616)
-0.646
(0.806)
1.003
(1.741)
-0.263
(1.595)
-3.293**
(1.360)
1.814
(1.307)
(3)
-0.085
(0.173)
0.002
(0.014)
-0.091***
(0.030)
3.218**
(1.328)
-1.036
(1.002)
1.908
(3.068)
-2.370
(2.206)
-2.000
(1.512)
3.859***
(1.314)
-0.442***
(0.100)
(4)
0.211
(0.129)
-0.014
(0.012)
0.039
(0.025)
-0.401
(0.640)
0.057
(0.713)
0.409
(1.310)
2.294*
(1.206)
-2.241**
(0.967)
-1.367
(1.106)
0.374***
(0.077)
(5)
0.204
(0.204)
-0.035*
(0.019)
0.048
(0.040)
-1.990**
(0.888)
0.864
(0.816)
-1.174
(3.196)
2.933*
(1.591)
-1.422
(1.336)
-1.820
(1.618)
(6)
0.000
(0.169)
-0.064***
(0.019)
-0.005
(0.026)
-1.893*
(0.986)
4.333***
(1.067)
2.278
(3.295)
7.163***
(1.769)
-6.210***
(1.533)
-7.544***
(1.584)
1.082***
(0.149)
-0.241
(0.624)
-1.221
(0.981)
-0.646
(0.806)
0.357
(1.435)
-1.036
(1.002)
0.872
(2.745)
0.057
(0.713)
0.466
(1.234)
0.864
(0.816)
-0.311
(3.148)
4.333***
(1.067)
6.611**
(3.202)
-1.029
(1.079)
-1.359*
(0.708)
0.477
(1.079)
0.148
(0.935)
-0.263
(1.595)
-3.556***
(1.114)
1.551
(1.065)
-1.742
(1.341)
-2.370
(2.206)
-4.369***
(1.273)
1.489
(2.086)
-0.511
(1.499)
2.294*
(1.206)
0.052
(0.841)
0.927
(1.087)
-1.314
(1.153)
2.933*
(1.591)
1.511
(1.416)
1.113
(1.486)
-0.309
(1.482)
7.163***
(1.769)
0.953
(0.888)
-0.382
(1.522)
-6.591***
(1.614)
165
-202.969
98
-120.432
93
-99.444
165
-185.635
98
-92.642
93
-71.386
X
X
X
X
X
X
X
X
Notes: Robust standard errors are reported in parentheses. Sample limited to cases with more than 5
jurors responding to school, income, age, race, gender, and religious questions. ∗=10%, ∗∗=5%, ∗ ∗ ∗=1%.
All columns also include number of counts; case, voir dire, and foreperson characteristics; and county and
case type fixed effects.
21
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
expected number of acquittals.38 Likewise, a higher proportion of religious individuals on
the jury is associated with a large increase in the expected number of convictions, and a
large decrease in the expected number of acquittals, with a 10-percentage point increase in
the proportion of religious jurors approximately associated with a 19 percent decrease in the
expected number of acquittals.39 These results directly correspond with the generally proprosecution bias we identified previously with religious and high-income individual jurors.
In contrast, some of the effects of jury composition on verdicts that we uncover do not
correspond as directly with relations between the associated individual juror characteristics
and leanings. Whereas jurors with more years of education are more likely to interpret
evidence in favor of the prosecution, juries with a higher average level of education convict
at a weakly lower rate and acquit at a weakly higher rate, though the coefficients are not
estimated precisely. On the other hand, whereas juror age was neither statistically nor
substantially associated with any individual juror leanings, the results in Table 4 indicate
that juries with a higher average age convict on significantly fewer counts. This latter finding
is notable because it stands in contrast to the findings of Anwar et al. (2012a), which suggest
that older jurors are more likely to convict than younger jurors. One possible explanation
for these differing conclusions about the effect of age is that Anwar et al. (2012a) cannot
control for juror characteristics beyond age, race, and gender. Since older individuals are
more likely to earn higher incomes, the age effect that Anwar et al. (2012a) find may be
biased by the negative impact of income on acquittals, as they are forced to omit income
from their analysis due to data limitations.40
As with our analysis of individual juror inclinations, it is difficult to draw firm conclusions
concerning effects of the gender composition of juries on trial outcomes. There is some
evidence that juries with a greater proportion of women are more lenient towards defendants
38
Poisson coefficients are semi-elasticities, i.e. proportional changes in the dependent variable for an
infinitessimal level change in the independent variable of interest. Of course, these are only approximate
when the contemplated change in the independent variable is large. Poisson coefficients can alternatively
be exponentiated to calculate what are known as the associated incidence-rate ratios, which give the exact
ratio of the level of the dependent variable corresponding to a unit increase in the independent variable of
interest to the level of the dependent variable corresponding to some baseline set of values of the regressors.
We discuss our results in terms of semi-elasticities in order to retain a transparent link to the estimated
coefficients reported in Table 4
39
Increasing the proportion of the jury with a given characteristic by ten percentage points (i.e. by 0.1) is
approximately equivalent to adding one person with that characteristic to a twelve-person jury.
40
It should be mentioned, however, that the research design of Anwar et al. (2012a) allows them to isolate
the effect of random variation in the age composition of juries on conviction rates. So while they are able to
deal with endogeneity in return for potentially suffering from a separate omitted variable issue, our ability
to include additional juror characteristics comes with a potential endogeneity issue that our proxy variable
approach may not fully address. Our analysis of individual juror biases, which includes case fixed effects in
addition to a full suite of juror characteristics, does not produce such directly contrasting results to the age
effect identified by Anwar et al. (2012a), though neither does it provide corroborating evidence.
22
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
in terms of convicting on fewer counts and acquitting on more counts. This in contrast to
the weak general bias in favor of the prosecution exhibited by individual female jurors that
we discussed above. However, the effect on trial outcomes is only statistically significant in
one specification. There are also signs that such leniency towards defendants is attenuated
when the victim is female, but again, the effects are weak and imprecisely estimated.
Finally, we find racial composition effects on trial outcomes to depend primarily on victim
race. In cases with a black victim, we find that a greater proportion of blacks on the jury
is associated with significant increases in the number of acquittals and decreases in the
number of convictions. The positive effect on the number of acquittals is most prominent
when the defendant is not black, while the negative impact on convictions is strongest when
the defendant is black. The effects are quite large, with a 10-percentage point increase
in the proportion of blacks associated with a decrease of between 10 and 44 percent in
convictions depending on the specification and the race of the defendant.41 These results
are consistent with the general pro-defense biases we noted amongst individual black jurors.
When the victim is not black, the effects are imprecisely estimated, and vary widely in sign
and magnitude across specifications.42
Although we find that many aspects of jury makeup are significant predictors of trial
outcomes, we find that more substantive aspects of the trial are also strong predictors of
verdicts. For example, the judge’s opinion regarding the evidence presented at trial as
favoring the defense versus the prosecution has a large impact on the number of convictions
and acquittals, in the expected direction.43
Even though we cannot know what the “true” or “correct” outcome should have been
for any of the trials in our dataset, we can use this measure of the strength of evidence to
put the magnitude of the jury composition effects in the context of factors that are closer
to core notions of justice. Suppose that there could never be evidence stronger than one
41
Note that, whereas the variable ‘black’ referred to an indicator of the individual juror’s race in Table 2,
it refers to the proportion of the jury that is black in Table 4.
42
Perhaps surprisingly, in a case with a black defendant and a non-black victim, our results from the
specification including all controls suggest that a greater proportion of blacks on the jury is associated with
a substantial and significant decrease in the number of acquittals. One possible explanation for this somewhat
surprising result may be that the victim not black indicator has a composite excluded category capturing
all cases with a black victim as well as cases without a victim, and that black jurors are only more likely to
acquit in cases without a victim. We do, however, control directly for whether there is a victim in the case,
which mitigates this concern; and specifications with alternative victim variable definitions and interactions
do not produce qualitatively different findings. A more plausible explanation is that the specification in
column 6 is over-fitting the data, which is supported by the small number of included trials with non-black
defendants and/or non-black victims and the corresponding large jumps in the magnitude and significance
of the race coefficients moving to this columng from the other specifications. We therefore do not regard this
particular estimate as reliable or indicative of a general effect.
43
While not shown in Table 4 in consideration of space, we also find that the judge’s assessments of other
characteristics of the trial and evidence are significant correlates of verdicts.
23
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
standard deviation below the mean strength of evidence against any defendant who is truly
innocent, and, symmetrically, that there could never be evidence weaker than one standard
deviation above the mean against a defendant who is truly guilty. Then, for a truly guilty
defendant to be acquitted, factors besides the strength of evidence would have to have an
effect equivalent to at least a two-standard-deviation shift in the strength of evidence. As
shown in Table 3, the sample standard deviation of the judge’s assessment of the strength
of evidence is 1.36, so that the estimates in Table 4 imply that a two-standard-deviation
shift in the strength of evidence in favor of the defendant would be associated with an 81120 percent decrease in the number of convictions per trial and a 102-294 percent increase
in the number of acquittals. To achieve the same effect through jury composition would
require, other things equal, a decrease in the average income of the jury by between $16,000
and $46,000, or an increase in the number of blacks on the jury of between 3 and 6. Or
equivalently, to make a defendant confronted with evidence of average strength appear truly
innocent or truly guilty, the race of roughly one third of the seated twelve-person jury would
have to be changed, or the average income of the jury would have to be changed by roughly
40% starting from the sample average jury income level.
An alternative way to place the jury composition effects in perspective is to employ
the regression-based technique proposed by Fields (2004) that decomposes the explanatory
power of of an overall model into the contributions of the constituent independent variables.
The basic idea is as follows. Let s(X k ) be the share of the variation in the dependent variable
Y that can be attributed to variation in the k-th independent variable. Fields defines s(X k )
as
s(X k ) ≡
cov(X k β̂ k , Y )
,
var(Y )
(2)
P
k
where β̂ k is the estimated coefficient on X k from our regression of choice. Then K
k=1 s(X ) =
R2 where K is the total number of independent variables in the model. We can then express
the s(X k ) in terms of their percentage contribution to the R2 of the regression model:
p(X k ) ≡
s(X k )
.
R2
(3)
Performing the Fields (2004) regression-based decomposition thus allows us to compare
the relative explanatory power of jury composition with that of evidentiary strength.44 For
44
In the poisson model, the traditional R2 is not well defined, so the decomposition must be performed
using the pseudo-R2 . An alternative would be to perform the decomposition for the analogous linear model.
This alternative is suggested by Fields (2004), though he states that the practical differences tend to be minor
for count data applications. We have likewise found little qualitative difference in results when applying the
24
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
convictions, variation in jury composition – in terms of education, income, age, religiousness,
sex, and race – accounts for about 39% of the explained variation in our model in column
3, about half of which is from the effect of race. However, roughly 43% of the explained
variation in convictions can be attributed to the set of variables associated with the judge’s
opinion of the evidentiary complexity and strength. For acquittals, 33% of the explained
variation can be attributed to jury composition, and the judge’s evaluation of the case’s legal
and evidentiary complexity and strength account for about 44% of the explained variation.
Therefore, while the proportion of variation in trial outcomes explained by jury composition
is not trivial, it is nonetheless exceeded by the proportion of variation explained by the more
fundamental aspects of fact and legal context.
4.3
Summary
We have demonstrated in this section that there is some basis for concerns that jury composition can affect whether justice is achieved in criminal trials. Opinions, sympathies, and
inclinations that would lead a juror to be biased towards one side are indeed correlated
with observable juror characteristics. And these individual biases do indeed appear to feed
through into verdicts when juries are composed of a larger number of individual jurors with
the associated observable characteristics.
However, it is important to note that the relevant observable characteristics are not
restricted to race, which has been the primary focus of several past studies. Most notably,
we also find income to be an important predictor of individual biases and determinant of trial
outcomes. While this is a novel and potentially important finding on its own, it also raises
the possibility that the race effects identified in past studies are confounded with effects that
are properly attributed to separate characteristics excluded from the analysis. Further, these
jury composition effects are smaller and have less explanatory power than effects associated
with the more fundamental aspects of the trial, i.e. those related to fact and the relevant
legal doctrine.
Nonetheless, concerns that jury composition effects – regardless of size or source – can
impede fairness in criminal trials remain. Furthermore, they raise a potential further concern,
which we now proceed to evaluate: whether attorneys are aware of these jury composition
effects, and might therefore attempt to make use of them to manipulate trial outcomes.
decomposition to a linear model.
25
Jury Composition, Trial Outcomes, & Attorney Preferences
5
Lehmann & Smith
Attorney Preferences Over Juror Characteristics
Motivated by our findings that there exist significant and robust relations between juror
characteristics – especially income, religiousness, and race – and both individual juror biases
and verdict patterns, we now turn to assessing whether attorneys anticipate these relations.
If attorneys are unaware of these effects, this may mitigate concerns of their impact on justice
(and would certainly mitigate concerns of endogeneity in our trial outcomes specifications).
If, however, attorneys are aware of these effects, we would expect them to prefer seated
juries with greater proportions of jurors exhibiting characteristics that we have found to be
amenable to their side. In order to analyze attorney preferences over jury composition and
assess if this is true, we capitalize on a self-reported measure of satisfaction concerning the
jury selection process that attorneys were requested to state before learning the verdict. In
the supplementary attorney survey, attorneys responded to the question, “How adequate was
the voir dire in this trial?” on a scale of 1 to 7, where 1 is “Very inadequate” and 7 is “Very
adequate.” We interpret higher responses on this scale as indicating that the responding
attorney felt better able to achieve a subjectively desirable jury composition.45 We then
ask whether this proxy measure for attorney satisfaction with the seated jury itself can be
explained by variation in the average characteristics of the seated jury, controlling for case
characteristics, various aspects of the voir dire process, attorney characteristics, and county
fixed effects.46
Before proceeding with the analysis, we take a closer look at the voir dire satisfaction
variables. Figure 1 illustrates the distribution across cases of the difference between the
defense attorney’s level of satisfaction and that of the prosecuting attorney. In about a third
of the cases, the defense and the prosecution report the same level of satisfaction, and in
another third of the cases, there is a one-point difference between the prosecution and the
defense. While one might expect to see a greater proportion of cases with more polarized
satisfaction, the shape of the distribution in Figure 1 could be driven by a few factors.
First, one strategy that both attorneys might follow is to exercise their strikes against the
45
In all of our analyses, we restrict our sample to trials for which the attorneys explicitly stated that they
had responded to the voir dire satisfaction question before learning the verdict. This supports our interpretation of these satisfaction ratings by making it less likely that they simply capture post hoc rationalization
of the broader trial outcome.
46
It is important to note that we do not attempt to examine the degree to which attorneys actually manage
to transform the composition of the jury in relation to that of the jury pool, nor do we attempt to identify the
precise actions taken or strategies employed by attorneys in voir dire. While these questions are interesting
in their own right, their answers depend both on attorney preferences and on a number of institutional and
trial-specific constraints, and it is the former that we wish to focus on exclusively here. In Lehmann and
Smith (2012), we develop a model of attorney behavior in voir dire to examine jury selection strategies and
outcomes in detail.
26
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Figure 1: Difference in Voir Dire Satisfaction within a Case: Defense Minus Prosecution
Notes: Sample limited to cases with more than 5 jurors responding to school, income, age, race, gender,
and religious questions and those cases in which both defense and prosecution answer voir dire satisfaction
question before learning the verdict.
potential jurors whom they perceive as having the most extreme prejudices against their
side. After all strikes have been exercised, most of those who are left on the seated jury
should then be jurors who are perceived as harboring only moderate inclinations to one side
or the other, and are therefore not associated with an extreme preference held by either
attorney. Second, if attorneys’ jury composition preferences were simply based on a single
characteristic of the jurors – for example, race – then one might expect a greater divergence
in satisfaction with the seated jury, since one side’s gain would necessarily be the other
side’s loss. However, if preferences are not unidimensional, then it is not clear that the
prosecution and the defense should express such antithetical satisfaction with the overall
makeup of the jury. For example, although the racial makeup of a given jury may benefit
the prosecution, the religiousness and income levels of the jurors might be simultaneously
more favorable to the defense. Finally, we do not know the benchmark against which any
given attorney might express his or her satisfaction. So, for example, in a trial in which the
jury pool has an unusually large proportion of blacks, the defense attorney might express a
high level of satisfaction for having a greater proportion of blacks on the seated jury than in
previous trials, while the prosecuting attorney might also express a high level of satisfaction
27
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
for successfully avoiding a seated jury with as many blacks as might have resulted from the
particular jury pool.
The empirical approach we take is to estimate separate ordered logistic models for defense
attorney satisfaction and prosecuting attorney satisfaction. This decision was based on a
careful evaluation of the strengths and weaknesses of alternative approaches, and on various
constraints imposed by our data.
The primary drawback of estimating separate models by attorney type is that it deprives
us of the ability to leverage the limited panel structure of our data to address unobserved caselevel heterogeneity. Since, in principle, we observe attorney satisfaction for each attorney
type within a case, we should be able to treat our dataset as a panel, with trial as the
cross-sectional dimension and two attorney-type observations for each cross-sectional unit.
In practice, however, there are unfortunately only 80-90 cases in our dataset for which both
attorneys responded to the voir dire satisfaction question before learning the verdict and for
which various other control variables are also available. The panel is hence unbalanced to a
substantial degree, though this in itself does not directly prevent us from employing a fixed
effects or related strategy. Rather, our main reason for avoiding panel methods is that they
render other important goals of our analysis difficult or impossible to achieve.
We would ideally like to account for the ordinal nature of the attorney satisfaction ratings,
which the ordered logistic form accomplishes. However, the ordered logistic model with fixed
effects leads to inconsistent estimates due to the incidental parameters problem. This can
be overcome by instead employing the consistent fixed effects ordered logistic estimator
proposed by Baetschmann et al. (2011), but the number of cases for which the associated
conditional likelihood function exists is limited by the unbalanced nature of the panel and by
the number of trials for which the prosecuting and defense attorneys reported the same values
for the dependent satisfaction variable. Taking this approach hence reduces our sample size
dramatically, leading to very imprecise estimates and susceptibility to over-fitting.
On the other hand, when we estimate separate linear specifications, implicitly treating the
dependent variables as continuous cardinal measures of attorney satisfaction, the results do
not differ substantially from the separate ordered logistic results that we present below. Since
this suggests that it might be relatively costless to simply ignore the non-cardinality of the
attorney satisfaction ratings, a seemingly natural way to proceed would be to pursue a linear
fixed effects approach. Such an approach would not give rise to the incidental parameters
problem, would not require a balanced panel – thus allowing us to retain as many observations
as possible – and is indeed the very approach we pursued in our analysis of individual juror
pre-deliberative biases. However, in the case of attorney satisfaction, a major drawback arises
with any of these fixed effects methods: since the explanatory variables of interest (namely,
28
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
average jury characteristics) vary only at the trial level, it becomes impossible to identify
the absolute effects of these variables on defense and prosecuting attorney satisfaction when
controlling for case fixed effects. Of course, by interacting the average jury characteristics
with a defense attorney indicator, we could estimate the effects of these characteristics on
defense attorney satisfaction relative to prosecutor satisfaction; but in addition to such
estimates having a less straightforward interpretation, they would not allow us to test our
hypotheses of interest. Having found, for example, that juries with higher average incomes
acquit on fewer counts, the natural prediction is that defense attorneys, if they are aware of
this effect, should, other things equal, be absolutely less satisfied with a jury with a higher
average income, not just relatively less satisfied than the prosecutor.
Our preferred choice in navigating these trade-offs is to focus on the separate ordered
logistic models for each side, which permits us to account for the ordinal nature of the
dependent variables, but more importantly, to identify the absolute effects of average jury
characteristics on the satisfaction of each attorney type. Moreover, this approach also allows
us to retain the greatest number of observations possible, thus leading to no disadvantage
in terms of the precision of the estimates. Unfortunately, this method leaves open the possibility that our estimates will be biased by unobserved case-level factors that simultaneously
affect attorney satisfaction and jury composition. As with our analysis of the effects of average jury characteristics on verdicts, we attempt to address such endogeneity concerns by
controlling for a rich set of trial-level covariates. Specifically, besides the jury characteristics
of primary interest and the characteristics of the attorney of the given side when applicable,
each specification also includes our full set of case and voir dire characteristics. In addition,
we have attempted to diagnose the presence of any remaining unobserved heterogeneity by
estimating alternative panel models – despite the difficulties just mentioned that arise with
such methods – and comparing the results to our baseline estimates. While the significance
and apparent magnitude of some effects vary widely across these alternative specifications,
the main results that we focus on below are relatively stable. We have therefore opted to
report our baseline results from the separate ordered logistic specifications only.47
47
The first set of alternative models combines the conditional binomial logit model with various methods
to collapse the satisfaction ratings from a seven-point to a two-point scale. Estimation of these models relies
on as few as 40 trials in some specifications. A somewhat more useful set of methods involves using the
satisfaction of one side to attempt to capture case fixed effects in specifications explaining the other side’s
satisfaction. One way in which we implemented this idea was to treat the satisfaction ratings as cardinal and
use their difference as the dependent variable. Alternatively, we included one side’s satisfaction directly in the
opposing side’s specification, entering as a set of dummy variables or simply linearly. Finally, we implemented
the linear fixed effects model with defense attorney interactions. We also made some attempts to increase
efficiency, first of all by pooling observations across attorneys and constraining some coefficients – including,
as in some other methods, the estimated logistic cut-off points on the unobserved latent satisfaction scale
corresponding to the observed categorical variables – to be the same for both sides (and without including case
29
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
These results are presented in Table 5. The first three columns examine the determinants
of the defense attorney’s satisfaction, and the last three columns report the correlates of the
prosecutor’s satisfaction. Columns 1 and 4 control for all variables summarized in Table 1
except for attorney and foreperson characteristics, which are excluded in order to retain as
many observations as possible. Specifications with these additional controls are reported in
the remaining columns, and demonstrate the robustness of our main results to these different
sets of controls.
The most striking finding in Table 5 is that the defense and prosecuting attorneys’ levels
of satisfaction are significantly correlated with the average income level of the seated jury,
in opposite directions and in line with our findings on individual biases and trial outcomes.
Seated juries with lower average income are associated with higher levels of satisfaction for the
defense, while the prosecution’s satisfaction level rises with a richer jury. This corresponds
directly with our findings of a general pro-prosecution bias amongst individual jurors with
higher incomes and the tendency for juries with higher average incomes to acquit on fewer
counts. The significance and robustness of the estimates provide a strong indication that
attorneys are well aware of these effects.
A second striking finding is that defense attorneys have a strong preference for seated
juries with a high proportion of women. This preference is most prominent when there is no
victim or a male victim, though the effect is still positive, if much less precisely estimated,
when there is a female victim. This result is puzzling in relation to the effects of juror
gender on individual biases and trial outcomes. While, as discussed above, juries with
a greater proportion of women do appear to be weakly more lenient towards defendants,
the effects are imprecisely estimated and somewhat sensitive to specification. Moreover,
individual female jurors appear to have at least a weak general bias towards the prosecution.
In contrast, the estimates in Table 5 indicate that defense attorneys have a very strong
preference for female jurors.
One potential explanation for this apparent divergence between our previous results and
attorney preferences is that attorneys use gender as an indicator of juror characteristics that
they have trouble observing directly. For example, attorneys might expect that women tend
to have lower incomes and fewer years of schooling or to be less religious, and – consistent
with the results in Table 2 showing that these traits are associated with a greater tendency to
interpret evidence in favor of the defense – prefer juries with a greater proportion of women
fixed effects); and, finally, by attempting to account for cross-attorney-equation correlation in a bivariate
probit model, again with collapsed binary satisfaction ratings. Each of these specifications is adversely
affected by the small sample sizes, whether through convergence problems, imprecise estimates, or nearly
exactly determined coefficients for some explanatory variables. However, the two main results that we focus
on are present and stable in sign and significance across all specifications.
30
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 5: Attorney Voir Dire Satisfaction: Ordered Logit Results
Dep. Var. = Attorney Satisfaction with Voir Dire (1 to 7 where 1=Low and 7=High)
Defense
Prosecution
education (years)
income ($, thousands)
age
religious
female
female × vfemale
black
black × dblack
black × vnblack
Female effect when
vfemale = 0
vfemale = 1
Black effect when
dblack = 0, vnblack = 0
dblack = 1, vnblack = 0
dblack = 0, vnblack = 1
dblack = 1, vnblack = 1
N
Log-Likelihood
Attorney Controls
Foreperson Controls
(1)
(2)
(3)
(4)
(5)
(6)
0.239
(0.204)
-0.046**
(0.018)
-0.017
(0.054)
1.548
(1.207)
3.678***
(1.111)
-0.203
(3.480)
1.026
(2.457)
-2.634
(2.480)
0.883
(2.491)
0.213
(0.218)
-0.054***
(0.020)
-0.018
(0.063)
1.851
(1.290)
4.183***
(1.209)
-0.062
(3.138)
2.085
(2.708)
-4.182
(2.780)
0.368
(2.868)
0.152
(0.242)
-0.060**
(0.023)
-0.006
(0.067)
0.703
(1.497)
4.684***
(1.438)
1.022
(3.499)
1.641
(3.120)
-4.401
(2.906)
1.353
(2.790)
-0.432*
(0.252)
0.067***
(0.025)
-0.001
(0.051)
0.555
(1.040)
0.784
(1.972)
-3.376
(3.195)
5.432*
(3.254)
-5.640**
(2.857)
-0.895
(2.077)
-0.360
(0.260)
0.076***
(0.028)
-0.000
(0.064)
0.955
(1.158)
-1.427
(2.004)
-3.664
(3.777)
6.096*
(3.161)
-6.371**
(2.770)
-1.227
(2.077)
-0.344
(0.299)
0.078**
(0.031)
0.017
(0.070)
0.465
(1.202)
-0.016
(2.017)
-7.122
(4.818)
6.331*
(3.720)
-5.862*
(3.232)
-1.451
(2.772)
3.678***
(1.111)
3.475
(3.268)
4.183***
(1.209)
4.121
(2.892)
4.684***
(1.438)
5.706*
(3.351)
0.784
(1.972)
-2.592
(2.874)
-1.427
(2.004)
-5.090
(3.387)
-0.016
(2.017)
-7.139*
(4.137)
1.026
(2.457)
-1.608
(1.314)
1.909
(2.097)
-0.725
(2.981)
2.085
(2.708)
-2.097
(1.390)
2.453
(2.409)
-1.729
(3.332)
1.641
(3.120)
-2.760*
(1.479)
2.993
(2.855)
-1.408
(3.046)
5.432*
(3.254)
-0.208
(1.490)
4.537*
(2.608)
-1.103
(2.108)
6.096*
(3.161)
-0.276
(1.598)
4.868*
(2.559)
-1.503
(2.091)
6.331*
(3.720)
0.469
(1.749)
4.880*
(2.947)
-0.983
(2.403)
142
-212.285
132
-186.467
X
124
-175.082
X
X
133
-203.166
122
-178.198
X
115
-167.010
X
X
Notes: Robust standard errors are reported in parentheses. Sample limited to cases with more than 5
jurors responding to school, income, age, race, gender, and religion questions and to cases in which
corresponding attorneys answered the voir dire satisfaction question before learning the verdict. ∗=10%,
∗∗=5%, ∗ ∗ ∗=1%. All models also control for county and case type fixed effects, and case and voir dire
characteristics.
31
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
on that basis. However, it appears that there would be little justification for attorneys
to form such beliefs. When we estimate the specifications in Tables 2 and 4 excluding
juror characteristics that attorneys may not be able to observe or approximate accurately,
the effects of juror gender are qualitatively unaffected.48 Thus, none of these alternative
specifications provides evidence that a greater proportion of women on seated juries, when
used as a proxy for other characteristics, would indicate any substantial advantage for the
defense.
A more straightforward intrepretation of this divergence between our previous results and
the strong preference of defense attorneys for female jurors may therefore be that attorneys
lack awareness of true relations between gender and individual biases or trial outcomes.
This might suggest that attorneys form preferences for the gender of seated jurors based on
anecdotal notions – even if they often turn out to be incorrect – that women are markedly
sympathetic to defendants or are very likely to possess other traits that are correlated with
such sympathies.49
Attorney difficulties with observing some juror characteristics may nonetheless be affecting our results through a different channel. In contrast to our findings of a strong association
between juror religiousness and pro-prosecution trial outcomes and individual juror biases,
attorneys do not appear to hold strong preferences for the average religiousness of the seated
jury. A possible explanation for the absence of a clear link between religiousness and attorney preferences is that the religiousness of a potential juror is less likely to be observable to
the attorneys than other characteristics. Hence, while attorneys may be aware of an effect
of religiousness on verdicts in general, they may have only a very limited notion either of the
religiousness of the seated jurors or the expected magnitude of the effect of their religiousness
in any particular case.50
48
Most notably, the strong tendency for female jurors to rate the prosecutor’s skill highly, from column 3
of Table 2, is unchanged in size and significance across every alternative specification. Similarly, the female
effect on the rating of the defense’s case becomes slightly smaller in absolute magnitude and loses significance
in some alternative specifications, but is always negative.
49
Table 5 also provides weak evidence that prosecuting attorneys prefer juries with a lower proportion
of females when there is a female victim. This might suggest that attorneys also rely to some extent on
anecdotal and frequently mistaken impressions that women are harsher judges of female victims.
50
One hypothesis arising from these conjectures is that religiousness may be more strongly correlated with
attorney satisfaction in cases in which attorneys were able to gather greater individual information about the
potential jurors through the use of questionnaires. We test this hypothesis by interacting the average level
of religiousness of the jury with a dummy indicating whether questionnaires were used in voir dire. In some
specifications, we find that, when a questionnaire was used, a higher proportion of religious jurors is weakly
associated with lower relative satisfaction for the defense attorney compared to the prosecution, broadly in
line with our findings on individual juror biases and trial outcomes. As mentioned in Section 2, regardless
of whether attorneys have access to results from supplementary questionnaires, they do know the basic
occupational and biographical details that courts routinely collect from all members of the jury pool upon
or prior to arrival at the courthouse. We therefore consider all of the other juror characteristics we observe
32
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Table 5 also suggests that prosecuting attorneys have a weak preference for juries with
fewer years of education on average, while defense attorneys may have a very weak preference for the opposite. This corresponds with the direction of the weak effects we found of
education on trial outcomes, though not with the association we found between individual
juror education and a greater tendency to interpret trial evidence in favor of the prosecution. Unfortunately, the imprecision of the estimates makes attorney preferences over juror
education difficult to analyze in more depth.
Finally, our results on attorney preferences for the proportion of blacks on the jury are
inconclusive and sometimes in conflict with what would be expected given our findings on the
impact of racial composition on verdicts. As discussed in Section 4, individual black jurors
exhibit strong pro-defense biases that are only slightly more prominent when the defendant
is black; while, in cases with a black victim, a higher proportion of blacks on the jury
is associated with more favorable trial outcomes for the defendant regardless of defendant
race. However, results in Table 5 suggest that the prosecutor holds some preference for
a greater proportion of black jurors when the defendant is not black, and also appear to
suggest that the defense may hold a weak preference for a less black-dominated jury when
the defendant is black, regardless of victim race. These results could perhaps be interpreted,
similar to those regarding juror gender, as indicating some degree of reliance on folk wisdom
amongst attorneys: prosecutors may believe that black jurors are unconditionally more likely
to be sympathetic towards black defendants, but simply not pay much attention to juror race
when the defendant is not black. However, inferences of this nature are difficult to support,
as the estimates of these race effects are imprecise, and their magnitude and significance are
sensitive to alternative specifications.
Another potential interpretation of these race results is that attorneys are expressing
their satisfaction relative to some unobserved benchmark: for example, as mentioned earlier,
a pool of mostly black potential jurors is likely to result in most of the seated jurors also
being black regardless of what the attorneys accomplish in jury selection, in which case
the prosecutor might be relieved and the defense attorney disappointed with any situation
except an all-black seated jury. It is also possible that, in response to increasing scrutiny
of race-based removal of potential jurors during jury selection, attorneys on both sides may
have moved away from treating racial composition as a predominant goal in jury selection
and instead formed stronger preferences for other juror characteristics.51 Our findings on
to also have been at least approximately known by the attorneys from their own observation and separate
information sources. Correspondingly, we find little impact of juror questionnaires on the effects of the other
juror characteristics on attorney preferences in alternative specifications with extended interactions.
51
As discussed in Section 2, striking jurors based solely on race is illegal, although there are practical
difficulties with enforcing this. We control for whether any Batson objections were raised in all specifications
33
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
the strong and robust effects of the jury’s average income and female representation on
attorney satisfaction certainly support the hypothesis that attorneys have multi-dimensional
preferences across characteristics besides race. It is thus possible that any estimated race
effects on attorney satisfaction are driven by a few isolated trials in our dataset, implying
that the best interpretation might be that these apparent race effects are just unreliably
estimated and masking a true zero effect.
In summary, we find evidence that attorneys are aware of some relations between juror
characteristics and trial outcomes. Of particular note are the strong and opposing preferences
over the average income of the jury held by attorneys on both sides, which indicate that the
ability to correctly anticipate the role of juror income in predicting juror biases and verdicts
is widespread in the legal profession. This serves to underscore the concerns that arise from
the existence of these relations in the first place, as it raises the possibility – or at least the
suspicion – that attorneys could use this knowledge to manipulate trial outcomes. At the
same time, our findings suggest that any attorney efforts in this regard may be hampered
by an imperfect understanding of the relation between some other juror characteristics and
trial outcomes and by an inability to observe some characteristics.
6
Conclusion
Our results give some justification for concerns about a lack of impartiality and trustworthiness in U.S. criminal jury trials. Defendants whose trials are heard by juries with a higher
average income and fewer blacks can expect the individual jurors to interpret the case and
evidence less in their favor and can expect a greater likelihood of conviction. Moreover,
as attorneys appear to have some ability to correctly anticipate these relations, defendants
must also worry that the prosecution will attempt to push the composition of the jury in
these specific directions. These findings hardly inspire confidence in the ability of the legal
system to deliver fair outcomes.
However, our findings also suggest some potential silver linings. First, our results do not
point to race as the primary source of bias in criminal trials. There is a popular perception
that attorneys predominantly target jurors for removal on the basis of race during jury
selection, resulting in widespread discriminatory exclusion of minorities from jury service
and a severe bias in verdicts against minority defendants. This perception may be primarily
driven by certain high-profile cases (such as the O.J. Simpson criminal trial or the first
in Table 5, but cannot observe which side raised such objections or what, if any, remedy was provided as
a result. We have attempted to test for the sensitivity of attorney preferences over race to the degree of
scrutiny concerning race-based strikes by interacting with the Batson indicator, but unfortunately the small
sample size prevents these effects from being estimated precisely enough to draw any conclusions.
34
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
trial of the officers who beat Rodney King), but it has also received some support from
various notable studies (especially Anwar et al., 2012b, Baldus et al., 2001, and Bowers et
al., 2001). Our results certainly confirm that black jurors tend to hold sentiments favorable
to defendants. On the other hand, our estimated effects of juror race on verdicts are less
pronounced than those on individual predispositions, and less pronounced as well than our
estimated effects of certain other juror characteristics on verdicts. Furthermore, our results
on attorney preferences in jury selection, while difficult to interpret, do not appear to indicate
that attorneys consider juror race to be associated with a substantial advantage to either
side. The difference in degree between our findings and those of other studies can likely be
explained by two important factors: the high and varying racial diversity in and across the
counties covered by our dataset; and the fact that we are able to jointly control for such a
breadth of juror characteristics. The robust income effects we find throughout our analysis
suggest that this may be an especially important omitted variable in other studies, leading to
an upward bias in the absolute magnitude of their estimated race effects due to the generally
greater population-wide propensity for minorities to have low incomes.
A second potential reason for optimism revealed by our results is that the substantive
evidentiary and legal aspects of trials remain the predominant determinants of verdicts. The
jury composition effects that we identify are small compared to the effect of evidentiary
strength, and account for less of the explained variation in verdicts than do the variables
related to the facts and legal doctrines central to the trial. Of course, the presence of
any significant jury composition effect could be interpreted as undermining the fairness and
validity of jury trials in general; and in a specific trial in which a defendant is truly innocent
but nonetheless faces moderately strong circumstantial evidence, even a small effect could
induce a false conviction if the prosecutor manages to alter the jury composition enough.
This question of how well attorneys can leverage jury composition effects to affect verdicts
is one that our present work only partially illuminates. To be able to do so at all, attorneys
must of course be aware of the jury composition effects in the first place, and this points
to a third potential silver lining of our results. While attorneys on both sides seem to be
well aware of the effect of juror income on trial outcomes, there are also signs that they are
at least partially mistaken about the effects of other characteristics. The defense’s strong
preference for women on seated juries, in particular, seems misguided in comparison to the
effects of juror gender on verdicts and individual biases that we estimate. And whether
attorneys are aware of the strong effects of religiousness or not, it appears that they have
trouble observing this trait in jurors.
But the ability of attorneys to affect trial outcomes through jury composition effects also
hinges on how much they can actually manage to manipulate jury composition in voir dire,
35
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
and our present work is silent on this issue. No matter how strong some individual juror
biases might be and how accurately attorneys anticipate them, the success of an attorney of a
given side in stacking the jury will be tempered by the actions and strategies of the opposing
counsel and by the institutional constraints imposed within the jury selection process. In
other words, the question of whether attorneys are aware of jury composition effects is moot
if they have no freedom to make use of such awareness in voir dire or if the actions of one
attorney can be perfectly offset by the other. This might suggest that a potentially attractive
approach to addressing some of the concerns arising from the effects of jury composition on
trial outcomes would be to simply limit attorney freedoms in voir dire. However, in Lehmann
and Smith (2012), we model attorney behavior in jury selection to evaluate these questions
in detail, and our findings suggests that such a policy may have its own drawbacks. In
particular, we find that giving attorneys more freedom in voir dire can be beneficial in some
cases in terms of allowing otherwise unobservable juror biases to be revealed. Having access
to this additional information can allow attorneys to break their reliance on stereotypes
based on observable juror characteristics like race and income, though trial outcomes will
nonetheless be affected if there are asymmetries across attorneys in how well each side can
access and use this information.
While we are therefore hesitant to make definitive policy recommendations at this stage,
our present findings do make an important contribution to jury and broader legal reform
debates by providing the most convincing evidence to date that the sources of jury bias run
deeper than race. Our results suggest that these sources are nuanced and multi-faceted,
encompassing several juror characteristics and attorney beliefs about them. The issue may
thus require an equally nuanced and multi-faceted policy response. Exploring how our results
generalize to regions and types of cases not covered by our dataset is a crucial next step in
eventually formulating such a response.
36
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
References
Abrams, David S., Marianne Bertrand, and Sendhil Mullainathan (2011) “Do Judges Vary
in their Treatment of Race?” Journal of Legal Studies, forthcoming.
Alesina, Alberto F. and Eliana La Ferrara (2011) “A Test of Racial Bias in Capital Sentencing,” National Bureau of Economic Research Working Paper, No. 16981.
American Civil Liberties Union of Northern California (2010) “Racial and Ethnic Disparities
in Alameda County Jury Pools,” Report.
Anwar, Shamena, Patrick Bayer, and Randi Hjalmarsson (2012a) “A Fair and Impartial
Jury? The Role of Age in Jury Selection and Trial Outcomes,” National Bureau of
Economic Research Working Paper, No. 17887.
(2012b) “The Impact of Jury Race in Criminal Trials,” Quarterly Journal of Economics, Vol. 127, pp. 1017–1055.
Baetschmann, Gregori, Kevin E. Staub, and Rainer Winkelmann (2011) “Consistent Estimation of the Fixed Effects Ordered Logit Model,” Institute for the Study of Labor
Discussion Paper, No. 5443.
Baldus, David C., George Woodworth, David Zuckerman, Neil Alan Weiner, and Barbara
Broffitt (2001) “The Use of Peremptory Challenges in Capital Murder Trials: A Legal
and Empirical Analysis,” University of Pennsylvania Journal of Constitutional Law,
Vol. 3, pp. 3–170.
Bowers, William J., Benjamin D. Steiner, and Marla Sandys (2001) “Death Sentencing in
Black and White: An Empirical Analysis of the Role of Jurors’ Race and Jury Racial
Composition,” University of Pennsylvania Journal of Constitutional Law, Vol. 3, pp.
171–274.
Diamond, Shari S., Destiny Peery, Francis J. Dolan, and Emily Dolan (2009) “Achieving
Diversity on the Jury: Jury Size and the Peremptory Challenge,” Journal of Empirical
Legal Studies, Vol. 6, pp. 425–449.
Equal Justice Initiative (2010) “Illegal Racial Discrimination in Jury Selection: A Continuing
Legacy,” Report.
Fields, Gary S. (2004) “Regression-Based Decompositions: A New Tool for Managerial
Decision-Making,” Cornell University Working Paper.
37
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Ford, Roger Allen (2010) “Modeling the Effects of Peremptory Challenges on Jury Selection
and Jury Verdicts,” George Mason Law Review, Vol. 17, pp. 377–381.
Gobert, James J., Ellen Kreitzberg, and Charles H. Rose (2009) Jury Selection: The Law,
Art, and Science of Selecting a Jury: West.
Greene, W. H. and D. A. Hensher (2010) Modeling Ordered Choices: A Primer: Cambridge
University Press.
Grosso, Catherine M. and Barbara O’Brien (2012) “A Stubborn Legacy: The Overwhelming
Importance of Race in Jury Selection in 173 Post-Batson North Carolina Capital
Trials,” Iowa Law Review, Vol. 97, pp. 1531–1559.
Hannaford-Agor, Paula L., Valerie P. Hans, Nicole L. Mott, and G. Thomas Munsterman
(2002) “Are Hung Juries a Problem?” National Center for State Courts Report.
(2003) “Evaluation of Hung Juries in Bronx County, New York, Los Angeles County,
California, Maricopa County, Arizona, and Washington, DC, 2000–2001,” National
Center for State Courts User Guide. Williamsburg, VA: National Center for State
Courts [producer]. Ann Arbor, MI: Interuniversity Consortium for Political and Social
Research [distributor].
Hoffman, Morris B. (2006) “Unnatural Selection,” New York Times, March 6.
Iyengar, Radha (2011) “Who’s the Fairest in the Land? Analysis of Judge and Jury Death
Penalty Decisions,” Journal of Law and Economics, Vol. 54, pp. 693–722.
King, Gary (1989) “A Seemingly Unrelated Poisson Regression Model,” Sociological Methods
and Research, Vol. 17, pp. 235–255.
Kressel, N. J. and D. F. Kressel (2002) Stack and Sway: The New Science of Jury Consulting:
Westview Press.
Lee, Jean (2010) “Do Jurors Discriminate? Evidence from State Juror Selection Procedures,”
5th Annual Conference on Empirical Legal Studies Paper.
Lehmann, Jee-Yeon K. and Jeremy Blair Smith (2012) “Attorney Empowerment in Voir Dire
and the Racial Composition of Juries,” Working Paper.
Lieberman, J. D. and B. D. Sales (2007) Scientific Jury Selection: American Psychological
Association.
38
Jury Composition, Trial Outcomes, & Attorney Preferences
Lehmann & Smith
Linder, Douglas (2007) “The Trials of Los Angeles Police Officers’ in Connection with the
Beating of Rodney King,” Social Science Research Network Working Paper.
Mize, Hon. Gregory E., Paula Hannaford-Agor, and Nicole L. Waters (2007) “The Stateof-the-States Survey of Jury Improvement Efforts: A Compendium Report,” National
Center for State Courts, Charlottesville, VA.
Myers, David G. and G. D. Bishop (1970) “Discussion Effects on Racial Attitude,” Science,
Vol. 169, pp. 778–779.
Neilson, William S. and Harold Winter (2000) “Bias and the Economics of Jury Selection,”
International Review of Law and Economics, Vol. 20, pp. 223–250.
Norton, M. I., S. R. Sommers, and S. Brauner (2007) “Bias in Jury Selection: Justifying
Prohibited Peremptory Challenges,” Journal of Behavioral Decision Making, Vol. 20,
pp. 467–479.
Pfeifer, J. E. (1990) “Reviewing the Empirical Evidence on Jury Racism: Findings of Discrimination or Discriminatory Findings?” Nebraska Law Review, Vol. 69, pp. 230–250.
Posey, A. J. and L. S. Wrightsman (2005) Trial Consulting: Oxford University Press.
Rose, M. R. (1999) “The Peremptory Challenge Accused of Race or Gender Discrimination?
Some Data from One County,” Law and Human Behavior, Vol. 23, No. 6, pp. 695–702.
Rothwax, Harold J. (1996) Guilty: The Collapse of Criminal Justice: Random House.
Shayo, Moses and Asaf Zussman (2011) “Judicial Ingroup Bias in the Shadow of Terrorism,”
Quarterly Journal of Economics, Vol. 126, pp. 1447–1484.
Sommers, Samuel R. (2008) “Determinants and Consequences of Jury Racial Diversity: Empirical Findings, Implications, and Directions for Future Research,” Social Issues and
Policy Review, Vol. 2, pp. 65–102.
Sommers, Samuel R. and Phoebe E. Ellsworth (2003) “How Much Do We Really Know about
Race and Juries? A Review of Social Science Theory and Research,” Chicago–Kent
Law Review, Vol. 78, pp. 997–1031.
Starr, V. Hale and Mark McCormick (2001) Jury Selection, Third Edition: Aspen Publishers
Online.
Sunstein, Cass R. (2000) “Deliberative Trouble? Why Groups Go to Extremes,” The Yale
Law Journal, Vol. 110, pp. 71–119.
39